6 ms·

Fujitsu’s Fugaku and A64FX Take Arm to the Top with 415 PetaFLOPs

66 points by dmmalam 6y ago

close04 6y ago
The thing that surprised me most is that when compared to Summit, the previous #1 powered by POWER9 cores and Nvidia GV100 accelerators, Fugaku seems to show the exact same power efficiency. > The new Fugaku supercomputer is bigger than Summit in practically every way. It has 3.05x cores, it has 2.8x the score in the official LINPACK tests, and consumes 2.8x the power So were the Nvidia accelerators what drove up Summit's efficiency?
- pedrocr 6y ago
  With the breakdown of Dennard scaling getting more performance per unit of energy is getting harder and harder. We need another breakthrough to get back on a clearly improving curve. It's surprising how it's exactly the same here though, given how different the two systems are.
- jabl 6y ago
  > So were the Nvidia accelerators what drove up Summit's efficiency? Yes. And the next generation NVIDIA accelerators (A100) in the Selene computer (#7 on the list) are even more efficient, currently taking the #2 spot on the green500 list (which is essentially the top500 list ordered by performance/watt). Edit: Disclaimer: I work for NVIDIA.
jabl 6y ago
See also closely related https://news.ycombinator.com/item?id=23601098 https://news.ycombinator.com/item?id=23601098 from yesterday.
mkl 6y ago
Did Apple convince these guys and TSMC (re 5nm production) to make their big ARM announcements the same day ARM Macs are announced? If it's not great marketing then it's a heck of a coincidence.
- jychang 6y ago
  Well, the WWDC rumors have been talking about ARM macs for weeks. If you were TSMC or Fujitsu and wanted to take advantage of a day where ARM was trending, you could schedule the announcement around WWDC without cooperating with Apple.
- davidmr 6y ago
  For the Top500 lists in particular, the list always comes out during the international supercomputing and supercomputing conferences in June and November respectively, so I’m sure there was no coordination. No idea about the TSMC thing, but I doubt it.
- IanCutress 6y ago
  TOP500 always comes out at ISC and SC. Whenever Apple does their announcements is up to them.
DCKing 6y ago
This year seems to be the year of big serious ARM performance. - AWS Graviton 2 is actually a viable option in the public cloud. - Apple starts transitioning their workhorses to ARM. - First place in the TOP500, apparently not due to custom magic or GPUs, but due to already existing/standardized ARM vector extensions. ARM is becoming such a household name that Fujitsu appears to have traded getting a paid ARM license over the royalty free SPARC instruction set they used to work with. I hope POWER can survive this onslaught, as SPARC surely has no hope of survival after this.
- lazyjones 6y ago
  Let's hope M$ will fix their OS to support ARM properly[1] soon, and that we get nice passive-cooled desktop systems. [1] no x64 support, limited OpenGL/3D support, no Fax/Scan support... AFAIK.
  - DCKing 6y ago
    For the Microsoft story I think we're also going to need some people actually making desktop worthy ARM chips. Qualcomm upping the power limit on their smartphone chips (as they are doing now) is not going to be enough to make Windows on ARM that attractive if you ask me. It doesn't help that ARM themselves are not confidently offering any desktop worthy intellectual property, as the smartphone chipmakers are just using ARM's own core designs.
    - test001only 6y ago
      I think Arm is actually starting to release design targeted for performance this year: https://www.anandtech.com/show/15813/arm-cortex-a78-cortex-x1-cpu-ip-diverging https://www.anandtech.com/show/15813/arm-cortex-a78-cortex-x...
      - DCKing 6y ago
        Yeah that's how ARM markets it, but if you read more carefully it's just a tweaked Cortex A78. A Cortex A78 is an impressive core design, but it's ultimately still a low power design. I'm saying the time is ripe for ARM to make a full fledged desktop focused design from the ground up, and if not ARM then someone else.
pjmlp 6y ago
And on that note, A64FX software stack also makes use of McKernel an an OS alternative. https://github.com/RIKEN-SysSoft/mckernel https://github.com/RIKEN-SysSoft/mckernel https://www.sys.r-ccs.riken.jp/ResearchTopics/os/mckernel/ https://www.sys.r-ccs.riken.jp/ResearchTopics/os/mckernel/
zapnuk 6y ago
How does it fare in terms of energy efficiency? Or is this not a concern at all?
- shenberg 6y ago
  Seems like energy efficiency is the same as the previous Top-1 (performance scaled linearly with the increase in power consumption), which is frankly not a great ad for the system.
  - jgtrosh 6y ago
    Setting as it has no accelerator, i'd say it shows how efficient the arm chips are, no?
newswasboring 6y ago
I remember back in university when I first read about the ARM business model, I naively thought it will always be the second fiddle to intel x86. At that time x86 had such a strong hold that for my young brain it was unimaginable that anyone else will capture the CPU market apart from embedded systems. The year 2020 is the conclusion of a long push by ARM and now I am wondering can RISC-V will be taking a similar path.
teruakohatu 6y ago
Are we getting to the point where we GPUs might be replaced by larabee style general purpose CPUs?
- pezezin 6y ago
  I seriously doubt it. GPUs are much more than the compute units; there is also the geometry setup, rasterizers, texture units, ROPs, etc. Doing all those tasks in software would be terribly expensive.
Merrill 6y ago
It appears that the Fujitsu A64FX is planned for Cray as well as its own A64FX machines, the PRIMEHPC FX1000 and PRIMEHPC FX700 models. https://www.hpcwire.com/2019/11/12/cray-fujitsu-both-bringing-fujitsu-a64fx-based-supercomputers-to-market-in-2020/ https://www.hpcwire.com/2019/11/12/cray-fujitsu-both-bringin... >The new HPE-Cray system, part of the Cray CS500 lineup, will employ the Fujitsu A64FX Arm-based processor with Arm Scalable Vector Extensions (SVE) and second-generation high-bandwidth memory (HBM). Named as customers in today’s release are Los Alamos National Laboratory, Oak Ridge National Laboratory, RIKEN Center for Computational Science, Stony Brook University, and University of Bristol. Cray and Fujitsu said they will be exploring engineering collaboration, co-development, and joint go-to-market strategies to meet customer demand as supercomputing extends into the exascale era.

66 points by dmmalam 6y ago

The thing that surprised me most is that when compared to Summit, the previous #1 powered by POWER9 cores and Nvidia GV100 accelerators, Fugaku seems to show the exact same power efficiency. > The new Fugaku supercomputer is bigger than Summit in practically every way. It has 3.05x cores, it has 2.8x the score in the official LINPACK tests, and consumes 2.8x the power So were the Nvidia accelerators what drove up Summit's efficiency?

pedrocr 6y ago
With the breakdown of Dennard scaling getting more performance per unit of energy is getting harder and harder. We need another breakthrough to get back on a clearly improving curve. It's surprising how it's exactly the same here though, given how different the two systems are.
jabl 6y ago
> So were the Nvidia accelerators what drove up Summit's efficiency? Yes. And the next generation NVIDIA accelerators (A100) in the Selene computer (#7 on the list) are even more efficient, currently taking the #2 spot on the green500 list (which is essentially the top500 list ordered by performance/watt). Edit: Disclaimer: I work for NVIDIA.

jabl 6y ago

See also closely related https://news.ycombinator.com/item?id=23601098 https://news.ycombinator.com/item?id=23601098 from yesterday.

mkl 6y ago

Did Apple convince these guys and TSMC (re 5nm production) to make their big ARM announcements the same day ARM Macs are announced? If it's not great marketing then it's a heck of a coincidence.

jychang 6y ago
Well, the WWDC rumors have been talking about ARM macs for weeks. If you were TSMC or Fujitsu and wanted to take advantage of a day where ARM was trending, you could schedule the announcement around WWDC without cooperating with Apple.
davidmr 6y ago
For the Top500 lists in particular, the list always comes out during the international supercomputing and supercomputing conferences in June and November respectively, so I’m sure there was no coordination. No idea about the TSMC thing, but I doubt it.
IanCutress 6y ago
TOP500 always comes out at ISC and SC. Whenever Apple does their announcements is up to them.

DCKing 6y ago

This year seems to be the year of big serious ARM performance. - AWS Graviton 2 is actually a viable option in the public cloud. - Apple starts transitioning their workhorses to ARM. - First place in the TOP500, apparently not due to custom magic or GPUs, but due to already existing/standardized ARM vector extensions. ARM is becoming such a household name that Fujitsu appears to have traded getting a paid ARM license over the royalty free SPARC instruction set they used to work with. I hope POWER can survive this onslaught, as SPARC surely has no hope of survival after this.

lazyjones 6y ago
Let's hope M$ will fix their OS to support ARM properly[1] soon, and that we get nice passive-cooled desktop systems. [1] no x64 support, limited OpenGL/3D support, no Fax/Scan support... AFAIK.
- DCKing 6y ago
  For the Microsoft story I think we're also going to need some people actually making desktop worthy ARM chips. Qualcomm upping the power limit on their smartphone chips (as they are doing now) is not going to be enough to make Windows on ARM that attractive if you ask me. It doesn't help that ARM themselves are not confidently offering any desktop worthy intellectual property, as the smartphone chipmakers are just using ARM's own core designs.
  - test001only 6y ago
    I think Arm is actually starting to release design targeted for performance this year: https://www.anandtech.com/show/15813/arm-cortex-a78-cortex-x1-cpu-ip-diverging https://www.anandtech.com/show/15813/arm-cortex-a78-cortex-x...
    - DCKing 6y ago
      Yeah that's how ARM markets it, but if you read more carefully it's just a tweaked Cortex A78. A Cortex A78 is an impressive core design, but it's ultimately still a low power design. I'm saying the time is ripe for ARM to make a full fledged desktop focused design from the ground up, and if not ARM then someone else.

pjmlp 6y ago

And on that note, A64FX software stack also makes use of McKernel an an OS alternative. https://github.com/RIKEN-SysSoft/mckernel https://github.com/RIKEN-SysSoft/mckernel https://www.sys.r-ccs.riken.jp/ResearchTopics/os/mckernel/ https://www.sys.r-ccs.riken.jp/ResearchTopics/os/mckernel/

zapnuk 6y ago

How does it fare in terms of energy efficiency? Or is this not a concern at all?

shenberg 6y ago
Seems like energy efficiency is the same as the previous Top-1 (performance scaled linearly with the increase in power consumption), which is frankly not a great ad for the system.
- jgtrosh 6y ago
  Setting as it has no accelerator, i'd say it shows how efficient the arm chips are, no?

newswasboring 6y ago

I remember back in university when I first read about the ARM business model, I naively thought it will always be the second fiddle to intel x86. At that time x86 had such a strong hold that for my young brain it was unimaginable that anyone else will capture the CPU market apart from embedded systems. The year 2020 is the conclusion of a long push by ARM and now I am wondering can RISC-V will be taking a similar path.

teruakohatu 6y ago

Are we getting to the point where we GPUs might be replaced by larabee style general purpose CPUs?

pezezin 6y ago
I seriously doubt it. GPUs are much more than the compute units; there is also the geometry setup, rasterizers, texture units, ROPs, etc. Doing all those tasks in software would be terribly expensive.

Merrill 6y ago

It appears that the Fujitsu A64FX is planned for Cray as well as its own A64FX machines, the PRIMEHPC FX1000 and PRIMEHPC FX700 models. https://www.hpcwire.com/2019/11/12/cray-fujitsu-both-bringing-fujitsu-a64fx-based-supercomputers-to-market-in-2020/ https://www.hpcwire.com/2019/11/12/cray-fujitsu-both-bringin... >The new HPE-Cray system, part of the Cray CS500 lineup, will employ the Fujitsu A64FX Arm-based processor with Arm Scalable Vector Extensions (SVE) and second-generation high-bandwidth memory (HBM). Named as customers in today’s release are Los Alamos National Laboratory, Oak Ridge National Laboratory, RIKEN Center for Computational Science, Stony Brook University, and University of Bristol. Cray and Fujitsu said they will be exploring engineering collaboration, co-development, and joint go-to-market strategies to meet customer demand as supercomputing extends into the exascale era.